xref: /openbmc/qemu/qemu-nbd.c (revision 1b111dc1)
1 /*
2  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
3  *
4  *  Network Block Device
5  *
6  *  This program is free software; you can redistribute it and/or modify
7  *  it under the terms of the GNU General Public License as published by
8  *  the Free Software Foundation; under version 2 of the License.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  *  GNU General Public License for more details.
14  *
15  *  You should have received a copy of the GNU General Public License
16  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
17  */
18 
19 #include "qemu-common.h"
20 #include "block/block.h"
21 #include "block/nbd.h"
22 #include "qemu/main-loop.h"
23 #include "block/snapshot.h"
24 
25 #include <stdarg.h>
26 #include <stdio.h>
27 #include <getopt.h>
28 #include <err.h>
29 #include <sys/types.h>
30 #include <sys/socket.h>
31 #include <netinet/in.h>
32 #include <netinet/tcp.h>
33 #include <arpa/inet.h>
34 #include <signal.h>
35 #include <libgen.h>
36 #include <pthread.h>
37 
38 #define SOCKET_PATH          "/var/lock/qemu-nbd-%s"
39 #define QEMU_NBD_OPT_CACHE   1
40 #define QEMU_NBD_OPT_AIO     2
41 #define QEMU_NBD_OPT_DISCARD 3
42 
43 static NBDExport *exp;
44 static int verbose;
45 static char *srcpath;
46 static char *sockpath;
47 static int persistent = 0;
48 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
49 static int shared = 1;
50 static int nb_fds;
51 
52 static void usage(const char *name)
53 {
54     (printf) (
55 "Usage: %s [OPTIONS] FILE\n"
56 "QEMU Disk Network Block Device Server\n"
57 "\n"
58 "  -h, --help           display this help and exit\n"
59 "  -V, --version        output version information and exit\n"
60 "\n"
61 "Connection properties:\n"
62 "  -p, --port=PORT      port to listen on (default `%d')\n"
63 "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
64 "  -k, --socket=PATH    path to the unix socket\n"
65 "                       (default '"SOCKET_PATH"')\n"
66 "  -e, --shared=NUM     device can be shared by NUM clients (default '1')\n"
67 "  -t, --persistent     don't exit on the last connection\n"
68 "  -v, --verbose        display extra debugging information\n"
69 "\n"
70 "Exposing part of the image:\n"
71 "  -o, --offset=OFFSET  offset into the image\n"
72 "  -P, --partition=NUM  only expose partition NUM\n"
73 "\n"
74 #ifdef __linux__
75 "Kernel NBD client support:\n"
76 "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
77 "  -d, --disconnect     disconnect the specified device\n"
78 "\n"
79 #endif
80 "\n"
81 "Block device options:\n"
82 "  -f, --format=FORMAT  set image format (raw, qcow2, ...)\n"
83 "  -r, --read-only      export read-only\n"
84 "  -s, --snapshot       use FILE as an external snapshot, create a temporary\n"
85 "                       file with backing_file=FILE, redirect the write to\n"
86 "                       the temporary one\n"
87 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
88 "                       load an internal snapshot inside FILE and export it\n"
89 "                       as an read-only device, SNAPSHOT_PARAM format is\n"
90 "                       'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
91 "                       '[ID_OR_NAME]'\n"
92 "  -n, --nocache        disable host cache\n"
93 "      --cache=MODE     set cache mode (none, writeback, ...)\n"
94 #ifdef CONFIG_LINUX_AIO
95 "      --aio=MODE       set AIO mode (native or threads)\n"
96 #endif
97 "\n"
98 "Report bugs to <qemu-devel@nongnu.org>\n"
99     , name, NBD_DEFAULT_PORT, "DEVICE");
100 }
101 
102 static void version(const char *name)
103 {
104     printf(
105 "%s version 0.0.1\n"
106 "Written by Anthony Liguori.\n"
107 "\n"
108 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
109 "This is free software; see the source for copying conditions.  There is NO\n"
110 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
111     , name);
112 }
113 
114 struct partition_record
115 {
116     uint8_t bootable;
117     uint8_t start_head;
118     uint32_t start_cylinder;
119     uint8_t start_sector;
120     uint8_t system;
121     uint8_t end_head;
122     uint8_t end_cylinder;
123     uint8_t end_sector;
124     uint32_t start_sector_abs;
125     uint32_t nb_sectors_abs;
126 };
127 
128 static void read_partition(uint8_t *p, struct partition_record *r)
129 {
130     r->bootable = p[0];
131     r->start_head = p[1];
132     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
133     r->start_sector = p[2] & 0x3f;
134     r->system = p[4];
135     r->end_head = p[5];
136     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
137     r->end_sector = p[6] & 0x3f;
138     r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
139     r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
140 }
141 
142 static int find_partition(BlockDriverState *bs, int partition,
143                           off_t *offset, off_t *size)
144 {
145     struct partition_record mbr[4];
146     uint8_t data[512];
147     int i;
148     int ext_partnum = 4;
149     int ret;
150 
151     if ((ret = bdrv_read(bs, 0, data, 1)) < 0) {
152         errno = -ret;
153         err(EXIT_FAILURE, "error while reading");
154     }
155 
156     if (data[510] != 0x55 || data[511] != 0xaa) {
157         return -EINVAL;
158     }
159 
160     for (i = 0; i < 4; i++) {
161         read_partition(&data[446 + 16 * i], &mbr[i]);
162 
163         if (!mbr[i].nb_sectors_abs)
164             continue;
165 
166         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
167             struct partition_record ext[4];
168             uint8_t data1[512];
169             int j;
170 
171             if ((ret = bdrv_read(bs, mbr[i].start_sector_abs, data1, 1)) < 0) {
172                 errno = -ret;
173                 err(EXIT_FAILURE, "error while reading");
174             }
175 
176             for (j = 0; j < 4; j++) {
177                 read_partition(&data1[446 + 16 * j], &ext[j]);
178                 if (!ext[j].nb_sectors_abs)
179                     continue;
180 
181                 if ((ext_partnum + j + 1) == partition) {
182                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
183                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
184                     return 0;
185                 }
186             }
187             ext_partnum += 4;
188         } else if ((i + 1) == partition) {
189             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
190             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
191             return 0;
192         }
193     }
194 
195     return -ENOENT;
196 }
197 
198 static void termsig_handler(int signum)
199 {
200     state = TERMINATE;
201     qemu_notify_event();
202 }
203 
204 static void *show_parts(void *arg)
205 {
206     char *device = arg;
207     int nbd;
208 
209     /* linux just needs an open() to trigger
210      * the partition table update
211      * but remember to load the module with max_part != 0 :
212      *     modprobe nbd max_part=63
213      */
214     nbd = open(device, O_RDWR);
215     if (nbd >= 0) {
216         close(nbd);
217     }
218     return NULL;
219 }
220 
221 static void *nbd_client_thread(void *arg)
222 {
223     char *device = arg;
224     off_t size;
225     size_t blocksize;
226     uint32_t nbdflags;
227     int fd, sock;
228     int ret;
229     pthread_t show_parts_thread;
230 
231     sock = unix_socket_outgoing(sockpath);
232     if (sock < 0) {
233         goto out;
234     }
235 
236     ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
237                                 &size, &blocksize);
238     if (ret < 0) {
239         goto out;
240     }
241 
242     fd = open(device, O_RDWR);
243     if (fd < 0) {
244         /* Linux-only, we can use %m in printf.  */
245         fprintf(stderr, "Failed to open %s: %m", device);
246         goto out;
247     }
248 
249     ret = nbd_init(fd, sock, nbdflags, size, blocksize);
250     if (ret < 0) {
251         goto out;
252     }
253 
254     /* update partition table */
255     pthread_create(&show_parts_thread, NULL, show_parts, device);
256 
257     if (verbose) {
258         fprintf(stderr, "NBD device %s is now connected to %s\n",
259                 device, srcpath);
260     } else {
261         /* Close stderr so that the qemu-nbd process exits.  */
262         dup2(STDOUT_FILENO, STDERR_FILENO);
263     }
264 
265     ret = nbd_client(fd);
266     if (ret) {
267         goto out;
268     }
269     close(fd);
270     kill(getpid(), SIGTERM);
271     return (void *) EXIT_SUCCESS;
272 
273 out:
274     kill(getpid(), SIGTERM);
275     return (void *) EXIT_FAILURE;
276 }
277 
278 static int nbd_can_accept(void *opaque)
279 {
280     return nb_fds < shared;
281 }
282 
283 static void nbd_export_closed(NBDExport *exp)
284 {
285     assert(state == TERMINATING);
286     state = TERMINATED;
287 }
288 
289 static void nbd_client_closed(NBDClient *client)
290 {
291     nb_fds--;
292     if (nb_fds == 0 && !persistent && state == RUNNING) {
293         state = TERMINATE;
294     }
295     qemu_notify_event();
296     nbd_client_put(client);
297 }
298 
299 static void nbd_accept(void *opaque)
300 {
301     int server_fd = (uintptr_t) opaque;
302     struct sockaddr_in addr;
303     socklen_t addr_len = sizeof(addr);
304 
305     int fd = accept(server_fd, (struct sockaddr *)&addr, &addr_len);
306     if (state >= TERMINATE) {
307         close(fd);
308         return;
309     }
310 
311     if (fd >= 0 && nbd_client_new(exp, fd, nbd_client_closed)) {
312         nb_fds++;
313     }
314 }
315 
316 int main(int argc, char **argv)
317 {
318     BlockDriverState *bs;
319     BlockDriver *drv;
320     off_t dev_offset = 0;
321     uint32_t nbdflags = 0;
322     bool disconnect = false;
323     const char *bindto = "0.0.0.0";
324     char *device = NULL;
325     int port = NBD_DEFAULT_PORT;
326     off_t fd_size;
327     QemuOpts *sn_opts = NULL;
328     const char *sn_id_or_name = NULL;
329     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:";
330     struct option lopt[] = {
331         { "help", 0, NULL, 'h' },
332         { "version", 0, NULL, 'V' },
333         { "bind", 1, NULL, 'b' },
334         { "port", 1, NULL, 'p' },
335         { "socket", 1, NULL, 'k' },
336         { "offset", 1, NULL, 'o' },
337         { "read-only", 0, NULL, 'r' },
338         { "partition", 1, NULL, 'P' },
339         { "connect", 1, NULL, 'c' },
340         { "disconnect", 0, NULL, 'd' },
341         { "snapshot", 0, NULL, 's' },
342         { "load-snapshot", 1, NULL, 'l' },
343         { "nocache", 0, NULL, 'n' },
344         { "cache", 1, NULL, QEMU_NBD_OPT_CACHE },
345 #ifdef CONFIG_LINUX_AIO
346         { "aio", 1, NULL, QEMU_NBD_OPT_AIO },
347 #endif
348         { "discard", 1, NULL, QEMU_NBD_OPT_DISCARD },
349         { "shared", 1, NULL, 'e' },
350         { "format", 1, NULL, 'f' },
351         { "persistent", 0, NULL, 't' },
352         { "verbose", 0, NULL, 'v' },
353         { NULL, 0, NULL, 0 }
354     };
355     int ch;
356     int opt_ind = 0;
357     int li;
358     char *end;
359     int flags = BDRV_O_RDWR;
360     int partition = -1;
361     int ret;
362     int fd;
363     bool seen_cache = false;
364     bool seen_discard = false;
365 #ifdef CONFIG_LINUX_AIO
366     bool seen_aio = false;
367 #endif
368     pthread_t client_thread;
369     const char *fmt = NULL;
370     Error *local_err = NULL;
371 
372     /* The client thread uses SIGTERM to interrupt the server.  A signal
373      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
374      */
375     struct sigaction sa_sigterm;
376     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
377     sa_sigterm.sa_handler = termsig_handler;
378     sigaction(SIGTERM, &sa_sigterm, NULL);
379 
380     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
381         switch (ch) {
382         case 's':
383             flags |= BDRV_O_SNAPSHOT;
384             break;
385         case 'n':
386             optarg = (char *) "none";
387             /* fallthrough */
388         case QEMU_NBD_OPT_CACHE:
389             if (seen_cache) {
390                 errx(EXIT_FAILURE, "-n and --cache can only be specified once");
391             }
392             seen_cache = true;
393             if (bdrv_parse_cache_flags(optarg, &flags) == -1) {
394                 errx(EXIT_FAILURE, "Invalid cache mode `%s'", optarg);
395             }
396             break;
397 #ifdef CONFIG_LINUX_AIO
398         case QEMU_NBD_OPT_AIO:
399             if (seen_aio) {
400                 errx(EXIT_FAILURE, "--aio can only be specified once");
401             }
402             seen_aio = true;
403             if (!strcmp(optarg, "native")) {
404                 flags |= BDRV_O_NATIVE_AIO;
405             } else if (!strcmp(optarg, "threads")) {
406                 /* this is the default */
407             } else {
408                errx(EXIT_FAILURE, "invalid aio mode `%s'", optarg);
409             }
410             break;
411 #endif
412         case QEMU_NBD_OPT_DISCARD:
413             if (seen_discard) {
414                 errx(EXIT_FAILURE, "--discard can only be specified once");
415             }
416             seen_discard = true;
417             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
418                 errx(EXIT_FAILURE, "Invalid discard mode `%s'", optarg);
419             }
420             break;
421         case 'b':
422             bindto = optarg;
423             break;
424         case 'p':
425             li = strtol(optarg, &end, 0);
426             if (*end) {
427                 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
428             }
429             if (li < 1 || li > 65535) {
430                 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
431             }
432             port = (uint16_t)li;
433             break;
434         case 'o':
435                 dev_offset = strtoll (optarg, &end, 0);
436             if (*end) {
437                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
438             }
439             if (dev_offset < 0) {
440                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
441             }
442             break;
443         case 'l':
444             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
445                 sn_opts = qemu_opts_parse(&internal_snapshot_opts, optarg, 0);
446                 if (!sn_opts) {
447                     errx(EXIT_FAILURE, "Failed in parsing snapshot param `%s'",
448                          optarg);
449                 }
450             } else {
451                 sn_id_or_name = optarg;
452             }
453             /* fall through */
454         case 'r':
455             nbdflags |= NBD_FLAG_READ_ONLY;
456             flags &= ~BDRV_O_RDWR;
457             break;
458         case 'P':
459             partition = strtol(optarg, &end, 0);
460             if (*end)
461                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
462             if (partition < 1 || partition > 8)
463                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
464             break;
465         case 'k':
466             sockpath = optarg;
467             if (sockpath[0] != '/')
468                 errx(EXIT_FAILURE, "socket path must be absolute\n");
469             break;
470         case 'd':
471             disconnect = true;
472             break;
473         case 'c':
474             device = optarg;
475             break;
476         case 'e':
477             shared = strtol(optarg, &end, 0);
478             if (*end) {
479                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
480             }
481             if (shared < 1) {
482                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
483             }
484             break;
485         case 'f':
486             fmt = optarg;
487             break;
488 	case 't':
489 	    persistent = 1;
490 	    break;
491         case 'v':
492             verbose = 1;
493             break;
494         case 'V':
495             version(argv[0]);
496             exit(0);
497             break;
498         case 'h':
499             usage(argv[0]);
500             exit(0);
501             break;
502         case '?':
503             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
504                  argv[0]);
505         }
506     }
507 
508     if ((argc - optind) != 1) {
509         errx(EXIT_FAILURE, "Invalid number of argument.\n"
510              "Try `%s --help' for more information.",
511              argv[0]);
512     }
513 
514     if (disconnect) {
515         fd = open(argv[optind], O_RDWR);
516         if (fd < 0) {
517             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
518         }
519         nbd_disconnect(fd);
520 
521         close(fd);
522 
523         printf("%s disconnected\n", argv[optind]);
524 
525 	return 0;
526     }
527 
528     if (device && !verbose) {
529         int stderr_fd[2];
530         pid_t pid;
531         int ret;
532 
533         if (qemu_pipe(stderr_fd) < 0) {
534             err(EXIT_FAILURE, "Error setting up communication pipe");
535         }
536 
537         /* Now daemonize, but keep a communication channel open to
538          * print errors and exit with the proper status code.
539          */
540         pid = fork();
541         if (pid == 0) {
542             close(stderr_fd[0]);
543             ret = qemu_daemon(1, 0);
544 
545             /* Temporarily redirect stderr to the parent's pipe...  */
546             dup2(stderr_fd[1], STDERR_FILENO);
547             if (ret < 0) {
548                 err(EXIT_FAILURE, "Failed to daemonize");
549             }
550 
551             /* ... close the descriptor we inherited and go on.  */
552             close(stderr_fd[1]);
553         } else {
554             bool errors = false;
555             char *buf;
556 
557             /* In the parent.  Print error messages from the child until
558              * it closes the pipe.
559              */
560             close(stderr_fd[1]);
561             buf = g_malloc(1024);
562             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
563                 errors = true;
564                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
565                 if (ret < 0) {
566                     exit(EXIT_FAILURE);
567                 }
568             }
569             if (ret < 0) {
570                 err(EXIT_FAILURE, "Cannot read from daemon");
571             }
572 
573             /* Usually the daemon should not print any message.
574              * Exit with zero status in that case.
575              */
576             exit(errors);
577         }
578     }
579 
580     if (device != NULL && sockpath == NULL) {
581         sockpath = g_malloc(128);
582         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
583     }
584 
585     qemu_init_main_loop();
586     bdrv_init();
587     atexit(bdrv_close_all);
588 
589     if (fmt) {
590         drv = bdrv_find_format(fmt);
591         if (!drv) {
592             errx(EXIT_FAILURE, "Unknown file format '%s'", fmt);
593         }
594     } else {
595         drv = NULL;
596     }
597 
598     bs = bdrv_new("hda");
599     srcpath = argv[optind];
600     ret = bdrv_open(bs, srcpath, NULL, flags, drv, &local_err);
601     if (ret < 0) {
602         errno = -ret;
603         err(EXIT_FAILURE, "Failed to bdrv_open '%s': %s", argv[optind],
604             error_get_pretty(local_err));
605     }
606 
607     if (sn_opts) {
608         ret = bdrv_snapshot_load_tmp(bs,
609                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
610                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
611                                      &local_err);
612     } else if (sn_id_or_name) {
613         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
614                                                    &local_err);
615     }
616     if (ret < 0) {
617         errno = -ret;
618         err(EXIT_FAILURE,
619             "Failed to load snapshot: %s",
620             error_get_pretty(local_err));
621     }
622 
623     fd_size = bdrv_getlength(bs);
624 
625     if (partition != -1) {
626         ret = find_partition(bs, partition, &dev_offset, &fd_size);
627         if (ret < 0) {
628             errno = -ret;
629             err(EXIT_FAILURE, "Could not find partition %d", partition);
630         }
631     }
632 
633     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed);
634 
635     if (sockpath) {
636         fd = unix_socket_incoming(sockpath);
637     } else {
638         fd = tcp_socket_incoming(bindto, port);
639     }
640 
641     if (fd < 0) {
642         return 1;
643     }
644 
645     if (device) {
646         int ret;
647 
648         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
649         if (ret != 0) {
650             errx(EXIT_FAILURE, "Failed to create client thread: %s",
651                  strerror(ret));
652         }
653     } else {
654         /* Shut up GCC warnings.  */
655         memset(&client_thread, 0, sizeof(client_thread));
656     }
657 
658     qemu_set_fd_handler2(fd, nbd_can_accept, nbd_accept, NULL,
659                          (void *)(uintptr_t)fd);
660 
661     /* now when the initialization is (almost) complete, chdir("/")
662      * to free any busy filesystems */
663     if (chdir("/") < 0) {
664         err(EXIT_FAILURE, "Could not chdir to root directory");
665     }
666 
667     state = RUNNING;
668     do {
669         main_loop_wait(false);
670         if (state == TERMINATE) {
671             state = TERMINATING;
672             nbd_export_close(exp);
673             nbd_export_put(exp);
674             exp = NULL;
675         }
676     } while (state != TERMINATED);
677 
678     bdrv_close(bs);
679     if (sockpath) {
680         unlink(sockpath);
681     }
682 
683     if (sn_opts) {
684         qemu_opts_del(sn_opts);
685     }
686 
687     if (device) {
688         void *ret;
689         pthread_join(client_thread, &ret);
690         exit(ret != NULL);
691     } else {
692         exit(EXIT_SUCCESS);
693     }
694 }
695